A while back I posted on a project called IKE (Inductive Knowledge Engine) that uses complexity theory to inductively organize digital reference knowledge bases. I’ve been back at work with it, doing a substantial rewrite of the underlying system. I’ve also spruced up the look of the IKE output (including a cloud view, but that’s for later). I thought I would share the most recent output.
In this figure 200 question/answer pairs from the AskVRD service are shown clustering. This is a very deterministic test run with 1 simple rule applied to all items…get closer to items that share your subject. This deterministic test is used to make sure all the software is working as expected. The next step is to include multiple rules, then dynamic rules (like how popular is this subject), and then finally diverse agent types. Still, I though the picture was pretty enough to share.
In case you want to know what the actual rule involved looks like (in PHP):
foreach ($prime as $p){
foreach ($target as $t){
if ($p[0]==$t[0]){
if (($p[1]==$t[1])AND($p[0]==”category”)){
$Move++;
}
if ($p[1]!=$t[1]){
//$Move–;
}
}
}
}